A novel general framework for evolutionary optimization: Adaptive fuzzy fitness granulation

Abstract

Computational complexity is a major challenge in evolutionary algorithms due to their need for repeated fitness function evaluations. Here, we aim to reduce number of fitness function evaluations by the use of fitness granulation via an adaptive fuzzy similarity analysis. In the proposed algorithm, an individual’s fitness is only computed if it has insufficient similarity to a queue of fuzzy granules whose fitness has already been computed. If an individual is sufficiently similar to a known fuzzy granule, then that granule’s fitness is used instead as a crude estimate. Otherwise, that individual is added to the queue as a new fuzzy granule. The queue size as well as each granule’s radius of influence is adaptive and will grow/shrink depending on the population fitness and the number of dissimilar granules. The proposed technique is applied to a set of 6 traditional optimization benchmarks that are for their various characteristics. In comparison with standard application of evolutionary algorithms, statistical analysis reveals that the proposed method will significantly decrease the number of fitness function evaluations while finding equally good or better solutions.